JMolMed DOI 10.1007/s00109-016-1466-4

ORIGINAL ARTICLE

Gene expression profiling at birth characterizing the preterm infant with early onset infection

Anne Hilgendorff1,2 & Anita Windhorst 3,4 & Manuel Klein5 & Svetlin Tchatalbachev3 & Christine Windemuth-Kieselbach6 & Joachim Kreuder2 & Matthias Heckmann7 & Anna Gkatzoflia2 & Harald Ehrhardt2 & Josef Mysliwietz8 & Michael Maier3 & Benjamin Izar3,9 & Andre Billion3 & Ludwig Gortner10 & Trinad Chakraborty3 & Hamid Hossain3

Received: 26 August 2015 /Revised: 9 August 2016 /Accepted: 18 August 2016 # The Author(s) 2016. This article is published with open access at Springerlink.com

Abstract with EOI could be further differentiated into two subclasses Early onset infection (EOI) in preterm infants <32 weeks ges- and were distinguished by the magnitude of the expression of tational age (GA) is associated with a high mortality rate and involved in both neutrophil and T cell activation. A the development of severe acute and long-term complications. hallmark activity for both subclasses of EOI was a common The pathophysiology of EOI is not fully understood and clin- suppression of genes involved in natural killer (NK) cell func- ical and laboratory signs of early onset infections in this pa- tion, which was independent from NK cell numbers. tient cohort are often not conclusive. Thus, the aim of this Significant results were recapitulated in an independent vali- study was to identify signatures characterizing preterm infants dation cohort. expression profiling may enable early and with EOI by using genome-wide (GWGE) more precise diagnosis of EOI in preterm infants. analyses from umbilical arterial blood of preterm infants. This prospective cohort study was conducted in preterm in- Key message fants <32 weeks GA. GWGE analyses using CodeLink hu- & Gene expression (GE) profiling at birth characterizes pre- man microarrays were performed from umbilical arterial term infants with EOI. blood of preterm infants with and without EOI. GWGE anal- & GE analysis indicates dysregulation of NK cell activity. yses revealed differential expression of 292 genes in preterm & NK cell activity at birth may be a useful marker to improve infants with EOI as compared to infants without EOI. Infants early diagnosis of EOI.

Anne Hilgendorff and Anita Windhorst contributed equally to the study. Electronic supplementary material The online version of this article (doi:10.1007/s00109-016-1466-4) contains supplementary material, which is available to authorized users.

* Hamid Hossain 4 Institute for Medical Informatics, Justus-Liebig-University Giessen, [email protected] Giessen, Germany 5 Hospital Barmherzige Brueder, Regensburg, Germany 1 Department of Neonatology, Grosshadern, Ludwig-Maximilian 6 Institute of Medical Biometry and Epidemiology, University Munich, Germany and the Comprehensive Pneumology Philipps-University Marburg, Marburg, Germany Center, Helmholtz Zentrum Muenchen, Munich, Germany, Member of the German Center for Lung Research (DZL), Munich, Germany 7 Department of Neonatology and Pediatric Intensive Care, University Medicine, Greifswald, Germany 2 Department of Pediatrics and Neonatology, Justus-Liebig University 8 Giessen, Germany, Member of the German Center for Lung Research Institute for Molecular Immunology, Helmholtz Center Munich, (DZL), Giessen, Germany Munich, Germany 9 Broad Institute of MIT and Harvard, Cambridge, MA, USA 3 Institute for Medical Microbiology, Justus-Liebig University Giessen, Member of the German Center for Infection Research 10 Department of Pediatrics and Neonatology, University of Saarland, (DZIF), Schubertstr. 81, 35392 Giessen, Germany Homburg, Germany J Mol Med

Keywords Neonatal sepsis . Gene expression profiling Patients were allocated to one of the two following groups: (I) EOI and (II) non-EOI (no signs of infection in the first 72 h of life). EOI was diagnosed if the infant showed both a pathologic Introduction ratio of immature to total granulocytes (IT ratio ≥ 0.2 as deter- mined by manual counts) and/or a pathologic white blood cell Preterm birth (PTB) remains a major problem in perinatal medi- count paralleled or followed by an increase in CRP ≥ 6mg/lin cine and accounts for 75 % of the perinatal mortality and 50 % of the first 72 h of life [7, 10–14]. These laboratory signs had to be the perinatal morbidity [1]. Early onset infections (EOI) in PTB are accompanied by at least three of the following clinical signs associated with significant morbidity and a mortality rate of 15– suggestive of bacterial infection in the new born infants: pallor, 50 %, especially in very immature preterm infants [2, 3]. Although gray skin color, capillary refill >3 s, requiring volume resuscita- clinical and experimental studies continue to provide valuable in- tion or substitution of any catecholamines, dyspnea, tachypnea, sight into processes leading to EOI and its attendant complications requiring respiratory support or supplemental oxygen, increased [4, 5], biological pathways relevant to the complex pathophysiol- thermal instability, unexplained hypo- and hyperglycemia, feed- ogy of EOI remain poorly understood. This is reflected in the ing difficulties, bilious reflux and abdominal distension, increas- limitations of early diagnostic markers and the absence of targeted ing incidence of apnea and/or bradycardia, lethargy, irritability, treatment approaches in this high-risk patient cohort. Clinical signs and increased or decreased muscle tone [2, 15]. All patients who are ambiguous and difficult to interpret in the absence of reliable did not meet the criteria for the EOI were considered as non-EOI. early biochemical markers [6–8]. The gold standard of blood The characteristics and clinical parameters in the first 72 h of life culture-proven sepsis severely underestimates the rate of severe of the patient cohort are given in Table 1 (exploration cohort). P infections in newborns and especially preterm infants, as the diag- values were calculated using Fisher’s exact test for qualitative nostic approach is complicated by maternal antibiotic therapy and parameters and Wilcoxon U test for quantitative parameters. hampered by small blood volumes [2, 9]. As outcome and prog- The comprehensive monitoring of the perinatal course is fur- nosis of EOI mainly depend on early and efficient treatment, sen- ther defined in Supplemental Materials and Methods. The study sitive and specific indicators of EOI are crucial at the earliest stage has been approved by the legal ethical committee (File 79/01, of disease. University of Giessen, Germany). In this prospective cohort study, we applied gene expres- sion profiling from umbilical arterial blood of preterm infants Blood sampling, RNA isolation, and microarrays <32 weeks of gestational age to provide comprehensive bio- logical information and identify biological pathways relevant Blood for standard laboratory analyses including WBC and for the development of EOI in order to enable the identifica- blood samples for transcriptome analyses were obtained from tion of early diagnostic markers. an indwelling umbilical artery catheter immediately after birth. WBCs were repeated upon clinical indication in the later postna- tal course as a possible indicator of developing (congenital and Material and methods nosocomial) infections. Details of the microarray experiments and data analysis can be found in Supplemental Materials and Patients Methods. Briefly, 250–300 μl of umbilical arterial blood was obtained Newborn infants <32 weeks gestational age (GA) were prospec- immediately after birth from an indwelling umbilical artery cath- tively included in this study. eter and directly transferred to 750–900 μl of the PAXgene Depending on the availability of diagnostic criteria at birth, Blood RNA System (PreAnalytiX, Heidelberg, Germany). infants were pro- or retrospectively excluded when one of the RNA isolation was performed according to the manufacturer’s following diagnoses was present: premature rupture of mem- recommendations (PreAnalytiX). RNA was hybridized on branes ≥3 weeks prior to birth leading to oligo- or anhydramnios, CodeLink UniSet Human 10 K Bioarrays (GE Healthcare) using severe congenital malformations, diagnosis of severe metabolic the CodeLink Expression Assay Kit (GE Healthcare) and sam- disorders, prepartum treatment of the mother with cytostatic or ples processed using CodeLink Expression Software V4.1 (GE immunosuppressive medication other than for lung maturation, Healthcare). and postnatal treatment with corticosteroids in a dose ≥1 mg/kg body weight. Analysis of C-reactive (CRP), whole white Gene expression analysis blood count (WBC), and microbiological examination of blood cultures, swabs, urine, and stool samples were carried out in the In order to account for confounding effects of WBCs on the first 72 h of life. Patients were clinically re-evaluated in short transcriptome pattern, we evaluated differences between EOI intervals and continuously monitored for vital signs, i.e., heart and non-EOI preterm infants in their differential WBCs at rate, blood pressure, microcirculation, and breathing pattern. birth by using the Wilcoxon -sum test. Missing data from JMolMed

Table 1 Neonatal characteristics of preterm infants EOI Non-EOI (control) P value

n 16 8 GA (weeks) 29 (24–30) 31 (29–31) 0.008 Birth weight (g) 1085 (590–1730) 1445 (900–1760) 0.046 IUGR 1 (6.25 %) 1 (12.25 %) 1 ANCS 6 (37.5 %)/a9 (56.3 %) 6 (75 %)/a6 (75 %) 0.667/a1 Chorioamnionitis 8 (50 %) 0 0.081 PROM 1 (6.25 %) 0 1 C-section 15 (93.75 %) 8 (100 %) 1 CRIB 6 (1–17) 1 (0–13) 0.102 RDS 16 (100 %) 7 (87.5 %) 0.333 RDS ≥ grade III 9 (56.3 %) 1 (12.5 %) 0.079 IVH 7 (43.75 %) 1 (12.5 %) 0.189 BPD 10 (62.5 %) 0 0.002 Length of mechanical ventilation (days) 7 (3–44) 0 (0–7) <0.001 ROP 9 (69 %) 4 (50 %) 0.646 Length of hospital stay (days) 70 (10–138) 47 (23–89) 0.065 Death 1 (6.25 %) 0 1 Blood culture positive 1 (6.25 %) 0 1 b b ITmax 0.32 (0.11–0.8) 0.17 (0.04–0.37) 0.018 b b CRPmax (mg/dl) 17.8 (7.1–52.3) 4(4–5.5) <0.001

Data are given as median and range or percent of total in group. P values are calculated using Fisher’s exact test for qualitative parameters and Wilcoxon U test for quantitative parameters EOI early onset infection, GA gestational age, IUGR intrauterine growth restriction, PROM premature rupture of membranes, ANCS antenatal corticosteroids: complete course including two doses of betamethasone given >24 h prior to birth, last dose <7 days before birth; CRIB critical risk index for babies, RDS respiratory distress syndrome, IVH intraventricular hemorrhage, BPD bronchopulmonary dysplasia, ROP retinopathy of prematurity a Any ANCS before birth b Value below lower determination threshold (4 mg/dl) was set to be 4 mg/dl

WBC counts resulting from technical problems or limited taken into account using limma in order to adjust their effect on sample size were imputed based on a model using a regular- gene expression analysis. Second, surrogate variable analysis ized iterative principal component analysis algorithm [16]tak- (SVA) [19] was conducted to account for hidden structures in ing into account relevant clinical data correlating with WBC, the cohorts, thereby excluding further unknown effects on gene i.e., GA, birth weight, maximum IT ratio, maximum CRP, expression analysis (for detailed description see Supplemental clinical risk index for babies (CRIB) score, and the presence Material and Methods). For SVA, two models were compared: of respiratory distress syndrome (RDS). the first model corrected gene expression analysis only for the The gene expression dataset was normalized using quantile effect of the aforementioned confounders; the second model addi- normalization in R [17]. For statistical analyses of the gene tionally took the EOI status into account. The identified surrogate expression data, a rank-based statistics, i.e., Rank Products, variable was used in limma to adjust gene expression analysis. was used to identify differentially regulated genes between Finally, Rank Products was used to analyze the adjusted data for EOI and non-EOI preterm infants. Being superior to classical differential gene expression. and moderated t statistics in studies with small sample sizes, Subsequent statistical analyses were conducted using the soft- this method was chosen for primary analysis [18]. A false ware tools BdChip^ for hierarchical clustering, BDAVID^ for gene discovery rate (FDR) was calculated for each transcript. ontology, and functional annotation clustering following the soft- To support the results derived from Rank Products and to ac- ware recommendations (Supplemental Materials and Methods). count for potential hidden confounders affecting gene expression analysis, the data were first corrected for variables that significantly Principal component analysis (PCA) correlated with structural differences between the groups, i.e., the EOI and non-EOI cohort (Table 1). These confounding variables, PCA as a mathematical vector space transformation allows for i.e., gestational age, birth weight, and WBCs, were subsequently the reduction of multidimensional data sets to lower J Mol Med dimensions (principle components) accounting for the vari- cell staining was performed with a PeCy5.5-labeled mouse ability of the data set [20]. PCA was conducted for 292 differ- IgG1 anti-human CD45 antibody from Invitrogen (Carlsbad, entially regulated genes as identified by Rank Products CA, USA) and separation of leukocyte fractions by simulta- analysis. neously using the following antibodies: PB-labeled mouse IgG1 anti-human CD3, Alx700-labeled mouse IgG1 anti- human CD19 and APC-labeled mouse IgG1 anti-human Disease load index (DLI) NKp46 from BD Biosciences (San Jose, CA, USA), APC- Alx750-labeled mouse IgG2a anti-human CD14 from To compare disease-dependent differences in the magnitude Invitrogen, FITC-labeled mouse IgG 1 anti-human CD15 of gene expression in preterm infants, we used an aggregate from Miltenyi Biotec (Bergisch Gladbach, Germany), and measure designated DLI as described previously [21]. The corresponding isotype controls. Flow cytometry was per- DLI is a unit-less measure representing the sum of the normal- formed on a LSRII Flow Cytometer (BD) using the proper ized expression values of defined differentially regulated controls to set gates and analyzed with FlowJo8.7.1 software. genes in an individual. Here, the DLI for each infant in clusters Dead cells were excluded using propidium iodide labeling and 1and2(Fig.1) was calculated. Subsequently, mean DLIs of doublets by gating on single cells (FSC-H to FSC-W channel). each patient group (non-EOI; EOI) were compared and the significant differences between the DLIs of each patient group were given as a P value derived from analysis of variances Validation and replication of the microarray results (ANOVA), pairwise Student’s t test with Benjamini-Hochberg by TaqMan RT-PCR correction for group A genes as well as the non-parametric Kruskal-Wallis test and the non-parametric pairwise Wilcoxon Microarray results were confirmed within the exploration co- ranks sum test with Benjamini-Hochberg correction for group hort and validated in an independent validation cohort by RT- B genes. The complete data set is available at the Gene PCR using TaqMan® technology (Applied Biosystems, Expression Omnibus (GEO) database under the accession Darmstadt, Germany) (Supplemental Materials and Methods). number GSE5760. Briefly, TaqMan quantitative real-time (RT)-PCR was per- formed for 10 human genes deriving from the microarray Measurement of NK cell number and activation results (ANXA1, CD163, GNLY, HIF1A, KLRC2, KLRD1, MPO, PGLYRP1, TNFRSF10A, CD177) and three house- Umbilical arterial blood specimens for measurement of natu- keeping genes glucose-6-phosphate dehydrogenase (G6PD), ral killer (NK) cell number were collected from a separate succinate dehydrogenase complex, subunit A, flavoprotein cohort of preterm infants (n = 20) < 32 weeks of GA included (Fp) (SDHA), and phosphoglycerate kinase 1 (PGK1) as in- and characterized exactly as described above. Hematopoietic ternal controls for normalization. To test whether the

Fig. 1 Transcriptional profiles of preterm using hierarchical clustering of and carbohydrate . Group B genes are downregulated in in- differentially expressed genes based on 292 differentially regulated genes. fants with EOI and mainly involved in NK cell activation. The group of Hierarchical clustering of differentially expressed genes of infants with infants with EOI could be further differentiated in a group with low and without EOI resulted in two main clusters (clusters 1 and 2). Group A expression of group A genes (EOI*) and EOI with high expression of genes are upregulated genes in infants with EOI and involved in group A genes (EOI**) neutrophil activation, T cell proliferation, hypoxia-induced signaling, JMolMed microarray results could be replicated, the gene expression of Table 2 White blood count of EOI and non-EOI the same 10 genes was investigated in a validation cohort White blood cells [×3/μl] EOI Non-EOI P value consisting of 43 new preterm infants by RT-PCR as described in Supplemental Materials and Methods. The patient charac- N Mean SE N Mean SE teristics of the validation cohort (15 patients with EOI, 28 non- EOI) are given in Supplemental Table 2. LEU 12 6.92 0.26 8 5.90 0.3 0.589 segNEU 11 3.44 0.38 6 1.29 0.15 0.145 bandNEU 11 2.26 0.28 5 0.30 0.05 0.141 Results juvNEU 10 0.38 0.03 7 0.37 0.03 0.695 LYM 11 3.48 0.16 6 4.42 0.29 0.248 Thirty very preterm infants were prospectively enrolled in this MON 11 0.96 0.12 6 0.57 0.05 0.960 study. Twenty-four of 30 samples met the high RNA quality P value from Wilcoxon U test criteria and were further processed for microarray analyses; of LEU leucocytes, segNEU segmented neutrophils, bandNEU band neutro- these, 16 were retrospectively allocated to the EOI cohort phils, juvNEU juvenile neutrophils, LYM lymphocytes, MON monocytes, based on the presence of clinical parameters for EOI in the SE standard error first 72 h of life. Eight infants without EOI were assigned to the control group (non-EOI). The characteristics of the patient cohorts are shown in Table 1. the cohort of preterm infants into clusters 1 and 2 (Fig. 1). Cluster 1 included preterm infants from both groups, EOI Gene expression analysis of umbilical arterial blood (EOI*), as well as non-EOI, while cluster 2 included infants reveals differential gene expression profiles in preterm with EOI (EOI**) with one exception. Thus, two subclasses infants with EOI at birth of EOI were identified, designated as EOI*, occurring mainly in cluster 1 and EOI** in cluster 2. The two subclasses of EOI Using a rank-based statistics, comparison of gene expression were also identified by PCA which provided a high degree of of infants with and without EOI revealed 292 differentially separation between the two subclasses EOI* and EOI** regulated genes (FDR ≤ 0.1). Of these, 219 genes had signif- (Fig. 2a, b). icantly higher gene expression levels (upregulated genes) in The two subclasses EOI* and EOI** were distinguished by infants with EOI, while 73 genes had significantly lower ex- the expression of two groups of genes, namely group A and B pression levels (downregulated genes) (Supplemental genes (Fig. 1): Group A genes were overexpressed in EOI** Table 1a, b). as compared to EOI* and non-EOI and were involved in neu- The differentially regulated transcripts were involved in pro- trophil activation, T cell proliferation, hypoxia-induced cesses related to inflammatory response (enrichment score signaling,andcarbohydrate metabolism. Group B genes were (ES) = 5.5), chemotaxis (ES 1.9), and leucocyte activation (ES downregulated in both EOI* and EOI** as compared to non- 1.8) as well as in catabolic processes (protein catabolic processes EOI and were involved in NK cell activation. To compare for ES = 1.8, ES = 1.2) (Supplemental Table 4). differences in the magnitude of the gene expression of group To account for unknown confounders and hidden struc- A and B genes, the aggregative DLIs of group A and B genes tures affecting gene expression analysis, the data set was in EOI* and EOI** were determined (Fig. 3): EOI** had a corrected for the variables GA, birth weight, and WBC significantly higher DLI for group A genes than both EOI* using SVA and limma. The analysis revealed a consider- (P = 0.0143) and non-EOI (P = 1e-05). Hence, the subclass able overlap of functional categories and the correspond- EOI** activated genes involved in neutrophil activation, T ing genes when compared with the results derived from cell proliferation, hypoxia-induced signaling, and carbohy- Rank Products only (Supplemental Tables 4 and 5), i.e., drate metabolism on a higher level than subclass EOI*. For the findings obtained from adjusted gene expression data group B genes, no significant differences occurred in the ex- supported the results from the initial Rank Products anal- pression level between the subclasses EOI* and EOI** ysis. Notably, comparison of the differential WBCs at (Fig. 3, P = 0.8292). But both EOI* and EOI** showed a birth showed no significant differences between EOI and significantly lower DLI for group B genes compared to non- non-EOI preterm infants (Table 2). EOI (P = 0.0280 and P = 0.0036, respectively) indicating decreased expression of genes involved in NK cell activation. Gene expression profiling reveals two groups of preterm Clinical variables characterizing EOI* and EOI** subclasses infants with EOI are given in Table 4, showing no statistically significant dif- ferences between the groups. However, although not signifi- Hierarchical clustering of the differentially regulated genes cant, preterm infants in the EOI* group showed more compli- (FDR ≤ 0.1) identified by Rank Products analysis separated cations than EOI**, i.e., higher RDS > grade III, J Mol Med

Fig. 2 a Three-dimensional principal component analysis (PCA) and b positions along the three axes derived from PCA. Patient boxplots of principal components (PC) based on 292 differentially subclassifications are indicated by color. PCA indicates a high degree regulated genes. Individual patients are plotted based on their respective of separation between the two subclasses EOI* and EOI** and non-EOI intraventricular hemorrhage (IVH), and development of neutrophil chemotaxis, adhesion, and migration in infants bronchopulmonary dysplasia (BPD). with EOI. Furthermore, increased expression of CD177, a receptor on the surface of neutrophils, indicated enhanced Increased activation of neutrophils in preterm infants transmigration. Strong activation of the neutrophils was also with EOI reflected by the overexpression of myeloperoxidase (MPO), neutrophil cytosolic factor 2 (NCF2), lactoferrin (LTF), Neutrophils play a pivotal role in the innate immune response, azurocidin 1 (AZU1), peptidoglycan recognition protein 1 as they migrate to the site of infection and help limit microbial (PGLYRP1)aswellascathepsinD(CTSD). infections. Increased activity of neutrophils in infants with EOI, especially in subclass EOI**, is indicated by overexpres- Decreased activation of NK cells in preterm infants sion of group A genes involved in phagocytotic activity, with EOI granula secretion, and respiratory burst of neutrophils as depicted in the interaction network in Fig. 4. The increased NK cells constitute a component of the innate immune system activation is given by overexpression of phospholipid in combating intracellular pathogens and activating and mod- scramblase 1 (PLSCR1), an enzyme involved in hematopoi- ulating the adaptive immune response. Activation of NK cells etic proliferation and differentiation of neutrophils (Table 3). is regulated by the expression of a variety of receptors. Overexpression of the proinflammatory calgranulins A In EOI, the NK cell activating killer cell lectin-like receptors (S100A8), B (S100A9), and C (S100A12) suggests enhanced (KLRs) such as KLR subfamily B member 1 (KLRB1), KLR

Fig. 3 Mean DLIs of differentially regulated group A and B genes in EOI* and EOI** compared to non-EOI. The significance of the preterm infants with and without EOI. Comparison of disease load indices difference between the DLIs of the patients groups was given as a P (DLIs) of non-EOI, EOI with low expression of group A genes (EOI*) value deriving from pairwise Student’s t test with Benjamini-Hochberg and EOI with high expression of group A genes (EOI**). Group A genes correction for group A genes and from non-parametric Kruskal-Wallis show significantly higher DLI of group A genes in EOI** compared to test and non-parametric pairwise Wilcoxon ranks sum test with EOI* and non-EOI. Group B genes show significantly lower DLI in both Benjamini-Hochberg correction for group B genes JMolMed

Fig. 4 Regulation of neutrophil activation. The gene interaction network activation, B binding, C causes/leads to, CC chemical–chemical regulation of neutrophil activation of the differentially regulated genes in interactions, CP chemical–protein interactions, E expression (includes EOI shows the interaction between calgranulins (S100A8, S100A9, metabolism/synthesis for chemicals), EC enzyme catalysis, I inhibition, S100A12) and genes involved in phagocytotic activity, granula L proteolysis (includes degradation for chemicals), LO localization, M secretion, and respiratory burst of neutrophils (e.g., MPO, AZU1, LTF, biochemical modification, MB group/complex membership, P NCF2). Upregulated genes are depicted in red and downregulated genes phosphorylation/dephosphorylation, PD protein–DNA interactions, PP in green. P value and fold change are given beneath each gene symbol. protein–protein interactions, PR protein–RNA interactions, RB Genes with an unknown regulation are depicted in white. Relationships regulation of binding, RE reaction, RR RNA-RNA interactions, T and interactions between molecules are abbreviated as follows: A transcription, TR translocation

subfamily C member 2 (KLRC2),andKLRsubfamilyDmem- infants with EOI could explain the reduced gene expres- ber 1 (KLRD1) were downregulated (Tables 3 and 4). GNLY, sion of NK cell receptors in these patients as depicted in whose expression is regulated via signaling through KLRB1, the interaction network in Fig. 5. KLRC2, and KLRD1, is an antimicrobial, cytolytic protein in The measurement of NK cell counts in a cohort of preterm the granules of NK cells and was also downregulated in EOI. infants with and without EOI showed no significant difference Two transcription factors, GATA-binding protein 3 in NK cell number between the two groups (non-EOI 7.8 ± 5.3 (GATA3) and CREB-binding protein (CREBBP), which cells/μl vs. EOI 5.8 ± 5.3 cells/μl; mean and standard devia- are known to regulate the expression of KLRs were down- tion (SD) each). This result suggested that the downregulation regulated. GATA3 is a transcription factor preferentially of NK cell activating genes as seen in EOI was not related to expressed in NK and T cells that plays an important role the NK cell count. in the early phase of NK cell development. Its activity is crucial for the diversification of the NK cell receptor rep- Validation of microarray results ertoire and interferon γ (IFNG) production and thus piv- otal for an effective NK cell response to viruses and bac- To validate the microarray data, TaqMan quantitative RT-PCR was teria. The downregulation of GATA3 and CREBBP in performed on 10 human target genes involved in EOI (ANXA1, J Mol Med

Table 3 Selected genes in relevant biological processes Biological process/gene name Symbol Fold FDR change

Neutrophil function Azurocidin 1 (cationic antimicrobial protein 37) AZU1a 1.57 0.001 Catalase CATa 3.51 0.000 Cathepsin D CTSD 2.00 0.007 CD177 CD177a 3.49 0.000 Formyl peptide receptor 1 FPR1 2.26 0.000 Grancalcin, EF-hand calcium binding protein GCAa 2.79 0.000 Lactotransferrin LTFa 2.82 0.000 Leukotriene A4 hydrolase LTA4H 2.34 0.003 Myeloperoxidase MPO 2.24 0.000 Neutrophil cytosolic factor 2 (65 kDa, chronic granulomatous disease, NCF2 2.08 0.004 autosomal 2) Peptidoglycan recognition protein 1 PGLYRP1a 2.42 0.000 Phospholipid scramblase 1 PLSCR1a 2.55 0.000 S100 calcium binding protein A8 (calgranulin A) S100 A8 2.07 0.001 S100 calcium binding protein A9 (calgranulin B) S100 A9 2.14 0.001 S100 calcium binding protein A12 (calgranulin C) S100 A12 2.42 0.000 S100 calcium binding protein P S100P 2.55 0.000 function Killer cell lectin-like receptor subfamily B, member 1 KLRB1 −1.75 0.069 Killer cell lectin-like receptor subfamily C, member 2 KLRC2a −1.64 0.030 Killer cell lectin-like receptor subfamily D, member 1 KLRD1 −1.83 0.053 Granulysin GNLYa −2.48 0.001 C-type lectin domain family 1, member B CLEC1B −1.70 0.036 Solute carrier family 30 (zinc transporter), member 1 SLC30A1 −1.89 0.071 Zinc finger, CCHC domain containing 2 ZCCHC2a −2.14 0.010 Zinc finger E-box binding homeobox 1 ZEB1 −1.72 0.007 Zinc finger protein 839 ZNF839 −1.72 0.089 Zinc finger protein 671 ZNF671a −2.69 0.000 Tcellfunction -related cell adhesion molecule 1 (biliary gly- CEACAM1 1.78 0.032 coprotein) Chemokine (C-X-C motif) receptor 4 CXCR4a 1.69 0.022 Hematopoietic cell-specific Lyn substrate 1 HCLS1a 2.21 0.002 Integrin, beta 1 (fibronectin receptor, beta polypeptide, antigen CD29 ITGB1 1.97 0.045 includes MDF2, MSK12) Interferon gamma receptor 2 (interferon gamma transducer 1) IFNGR2a 1.88 0.004 Interleukin 10 IL10 1.71 0.022 Protein tyrosine phosphatase, non-receptor type 22 (lymphoid) PTPN22 2.33 0.003 Protein tyrosine phosphatase, receptor type, C PTPRCa 2.30 0.001 Transcription Hypoxia-inducible factor 1, alpha subunit (basic helix-loop-helix tran- HIF1A 1.75 0.055 scription factor) CCAAT/enhancer binding protein (C/EBP), beta CEBPBa 2.35 0.000 CCAAT/enhancer binding protein (C/EBP), alpha CEBPA 2.10 0.002 B-cell CLL/lymphoma 6 BCL6 2.00 0.017 CREB binding protein CREBBP −1.95 0.023 GATA binding protein 3 GATA3a −1.84 0.043

a Significant in Rank Products and SVA JMolMed

Table 4 Characteristics of preterm infants EOI* and EOI** EOI* EOI** P value

n 511 GA (weeks) 28 (24–30) 29 (24–30) 0.910 Birth weight (g) 1060 (700–1390) 1100 (590–1730) 0.610 IUGR 0 1 (9 %) 1 ANCS 3 (60 %)/a4 (80 %) 3 (27 %)/a5 (55 %) 0.580/a0.228 Chorioamnionitis 2 (40 %) 6 (55 %) 1 PROM011 C-section 5 (100 %) 10 (91 %) 1 CRIB 2 (1–11) 7 (1–17) 0.351 RDS 5 (100 %) 11 (100 %) 1 RDS ≥ grade III 4 (80 %) 6 (55 %) 0.588 IVH 3 (60 %) 4 (36 %) 0.596 BPD 4 (80 %) 6 (55 %) 0.600 Length of mechanical ventilation (days) 7 (5–44) 7 (3–23) 0.597 ROP 2 (50 %) 7 (78 %) 0.596 Length of hospital stay (days) 74 (51–138) 70 (10–124) 0.844 Blood culture positive 1 (20 %) 0 0.267 b b ITmax 0.2 (0.1–0.7) 0.4 (0.1–0.8) 0.733 b b CRPmax (mg/dl) 12.4 (7.7–24.3) 18.9 (7.1–52.3) 0.257 White blood cell counts, mean (standard error) LEU 8.32 (0.97) 6.21 (0.32) 0.552 segNEU 3.37 (0.9) 3.46 (0.59) 0.759 bandNEU 4.02 (1.73) 1.6 (0.26) 0.475 juvNEU 0.43 (0.07) 0.35 (0.06) 0.594 LYM 3.32 (0.49) 3.54 (0.24) 1 MON 2.15 (0.77) 0.52 (0.05) 0.126

Data are given as median and range or percent of total in group. P values are calculated using Wilcoxon U test for quantitative parameters and Fisher’s exact test for qualitative parameters EOI early onset infection, GA gestational age, IUGR intrauterine growth restriction, ANCS antenatal corticoste- roids: complete course including two doses of betamethasone given >24 h prior to birth, last dose <7 days before birth, CRIB critical risk index for babies, RDS respiratory distress syndrome, IVH intraventricular hemorrhage, BPD bronchopulmonary dysplasia, ROP retinopathy of prematurity, LEU leucocytes, segNEU segmented neu- trophils, bandNEU band neutrophils, juvNEU juvenile neutrophils, LYM lymphocytes, MON monocytes a Any ANCS before birth b Value below lower determination threshold (4 mg/dl) was set to be 4 mg/dl

CD163, GNLY, HIF1A, KLRC2, KLRD1, MPO, PGLYRP1, The RT-PCR results in the validation cohort confirmed signif- TNFRSF10A, CD177). The overall correspondence between icant upregulation of ANXA1, CD163, MPO, PGLYRP1, mRNA levels measured by microarrays and by RT-PCR analyses HIF1A, TNFRSF10A, and CD177 in the group of infants with was high (R2 = 0.88) (Supplemental Fig. 1). EOI. In contrast, genes involved in NK cell activation, i.e., KLRC2, KLRD1, and GNLY, were found to be significantly downregulated (Fig. 6). Replication of the results in an independent validation The results show that the validation cohort could support the cohort key biologic findings found in the exploration cohort.

To test the reproducibility of the obtained results, 10 selected genes were analyzed by RT-PCR in a validation cohort (n = 43, Supplemental Table 2). The results show an overall good correla- Discussion and conclusions tion (R2 = 0.74) of the gene expression between the exploration and the validation cohort of preterm infants indicating reproduc- We performed transcriptional profiling from umbilical arterial ibility of the results (Supplemental Fig. 2). blood samples to obtain insights into the pathways involved in J Mol Med

Fig. 5 Influence of GATA3 and IL10 on regulation of NK cells and regulated genes in EOI shows the interaction of GATA3, IL10, and lymphocytes. The gene interaction network influence of GATA3 and CREBBP with KLRs. Relationships and interactions between IL10 on regulation of NK cells and lymphocytes of the differentially molecules are given in the figure legends of Fig. 4

Fig. 6 Comparison of gene expression between EOI and non-EOI for downregulation of genes involved in NK cell activation, i.e., KLRC2, selected genes in a second patient cohort (replication cohort). TaqMan KLRD1, and GNLY in the group of infants with EOI. P values are quantitative RT-PCR results of 10 selected genes were compared in given for ANXA1, PGLYRP1, and TNFRSF10A using the Welch test, infants with and without EOI within the replication cohort: RT-PCR for CD163, HIF1A, and GNLYusing the parametric Kruskal-Wallis rank- results confirmed a significant overexpression of ANXA1, CD163, sum test, and for KLRC2, KLRD1, MPO, and CD177 using the Student’s MPO, PGLYRP1, HIF1A, TNFRSF10A, and CD177 and a significant t test JMolMed early EOI development in very premature infants. Comparison of NK cell activation but differing in the activation of neutrophils, T the gene expression profiles of infants with EOI and without EOI cell proliferation, hypoxia-induced signaling, and carbohydrate revealed NK cell inactivation to be a hallmark of EOI discrimi- metabolism. nating EOI and non-EOI neonates (Figs. 1 and 2). We propose that the addition of NK cell activity into the Impairment of NK cell function plays a critical role in the standard diagnostic repertoire for critically ill (preterm) neo- host response to infectious challenges in preterm infants with nates could be a useful complement to current laboratory di- EOI, but a comprehensive evaluation of their cell state is lack- agnostics in order to improve early diagnosis of EOI. ing [22]. Our results indicate that decreased activation through downregulation of key surface markers, their regulating tran- Acknowledgments We thank Juri Schklarenko and Felix Thierer for scription factors GATA3 and CREBBP and downstream lytic excellent technical assistance. The present study was supported by the National Genome Network (NGFN), Germany (NGFN IE-S08T03) and enzymes account for the impairment of NK cell function, by the Pneumonia Research Network on Genetic Resistance and while NK cell numbers were similar among patients with Susceptibility for the Evolution of Severe Sepsis (PROGRESS), grant and without EOI. NK cell interactions with other immune number 01KI07110. cells are regulating a wide range of immune responses [23] including bacterial clearance during bacterial sepsis by NK Compliance with ethical standards cell and macrophage interaction [24]. The decrease in NK cell Funding The present study was supported by the National Genome activation may contribute to impaired clearance of pathogens Network (NGFN), Germany (NGFN IE-S08T03) and by the leading to overwhelming systemic infections, frequent in pre- Pneumonia Research Network on Genetic Resistance and Susceptibility mature infants. Insufficient elimination of pathogens due to for the Evolution of Severe Sepsis (PROGRESS), grant number impaired orchestration by NK cells could also explain the 01KI07110. excessive neutrophil response in patients with EOI, consistent Conflict of interest The authors declare no conflict of interests related with findings from studies in infants with fetal inflammatory to this study. response syndrome (FIRS) [25]. Recent studies in infants up to 3 years of age revealed impaired adaptive immune system responses and specifically inhibition of Open Access This article is distributed under the terms of the Creative NK cell activation in septic shock, supporting a key observation in Commons Attribution 4.0 International License (http:// our study [26, 27]. This is consistent with the findings of El- creativecommons.org/licenses/by/4.0/), which permits unrestricted use, Sameea et al. [28] and Georgeson et al. [29], who showed a distribution, and reproduction in any medium, provided you give positive correlation between reduced NK cell activity and the pres- appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. ence, severity, and outcome of neonatal sepsis in term newborns. Here, we show that similar concepts apply to EOI in pre- References term infants and validated the expression of key markers of NK cell activation (and the inhibition thereof) in an indepen- 1. Slattery MM, Morrison JJ (2002) Preterm delivery. Lancet 360: dent cohort. Overall, these findings suggest that a qualitative 1489–1497 NK cell impairment, measured by gene expression profiling, 2. Stoll BJ, Gordon T, Korones SB, Shankaran S, Tyson JE, Bauer may be inherent to EOI and provide a superior diagnostic tool CR, Fanaroff AA, Lemons JA, Donovan EF, Oh W, et al. (1996) for the early detection of EOI. Early-onset sepsis in very low birth weight neonates: a report from We identified two subclasses within the EOI cohort (EOI* the National Institute of Child Health and Human Development Neonatal Research Network. J Pediatr 129:72–80 and EOI**) characterized by increased neutrophil activation, T 3. Stoll BJ, Hansen NI, Sanchez PJ, Faix RG, Poindexter BB, Van cell proliferation, hypoxia-induced signaling, and carbohydrate Meurs KP, Bizzarro MJ, Goldberg RN, ID F 3rd, Hale EC, et al. metabolism in EOI** preterm infants. Clinically, the group of (2011) Early onset neonatal sepsis: the burden of group B strepto- – infants classified as EOI* showed more infection-associated coccal and E. coli disease continues. Pediatrics 127:817 826 4. Jobe AH (2003) Antenatal factors and the development of complications when compared with EOI**. In the of im- bronchopulmonary dysplasia. Semin Neonatol 8:9–17 paired NK cell activation in both EOI subclasses, increased acti- 5. Leviton A, Kuban K, O’Shea TM, Paneth N, Fichorova R, Allred EN, vation of neutrophils, T cell proliferation, hypoxia-induced sig- Dammann O (2011) The relationship between early concentrations of naling, and might indicate protective 25 blood and cerebral white matter injury in preterm new- borns: the ELGAN study. J Pediatr 158:897–903.e891-895 mechanisms helping to prevent complications of EOI in preterm 6. Laborada G, Rego M, Jain A, Guliano M, Stavola J, Ballabh P, infants when classified as EOI**. Krauss AN, Auld PA, Nesin M (2003) Diagnostic value of cyto- To conclude, the results of this study indicate that (1) tran- kines and C-reactive protein in the first 24 hours of neonatal sepsis. scriptome patterns derived from umbilical arterial blood enable Am J Perinatol 20:491–501 7. Mathers NJ, Pohlandt F (1987) Diagnostic audit of C-reactive pro- the discrimination of preterm infants with and without EOI and tein in neonatal infection. Eur J Pediatr 146:147–151 that (2) the group of preterm infants with EOI can be divided in 8. Ottolini MC, Lundgren K, Mirkinson LJ, Cason S, Ottolini MG two subclasses showing a common phenomenon of dysregulated (2003) Utility of complete blood count and blood culture screening J Mol Med

to diagnose neonatal sepsis in the asymptomatic at risk newborn. 20. Daffertshofer A, Lamoth CJ, Meijer OG, Beek PJ (2004) PCA in Pediatr Infect Dis J 22:430–434 studying coordination and variability: a tutorial. Clin Biomech 9. Connell TG, Rele M, Cowley D, Buttery JP, Curtis N (2007) How (Bristol, Avon) 19:415–428 reliable is a negative blood culture result? Volume of blood submit- 21. Porter JD, Merriam AP, Leahy P, Gong B, Khanna S (2003) ted for culture in routine practice in a children’s hospital. Pediatrics Dissection of temporal gene expression signatures of affected and 119:891–896 spared muscle groups in dystrophin-deficient (mdx) mice. Hum 10. Berger C, Uehlinger J, Ghelfi D, Blau N, Fanconi S (1995) Mol Genet 12:1813–1821 Comparison of C-reactive protein and white blood cell count with 22. Marodi L (2006) Neonatal innate immunity to infectious agents. differential in neonates at risk for septicaemia. Eur J Pediatr 154: Infect Immun 74:1999–2006 – 138 144 23. Zhang C, Zhang J, Tian Z (2006) The regulatory effect of natural 11. Rodwell RL, Taylor KM, Tudehope DI, Gray PH (1993) killer cells: do "NK-reg cells" exist? Cell Mol Immunol 3:241–254 Hematologic scoring system in early diagnosis of sepsis in neutro- – 24. Scott MJ, Hoth JJ, Gardner SA, Peyton JC, Cheadle WG (2003) penic newborns. Pediatr Infect Dis J 12:372 376 Natural killer cell activation primes macrophages to clear bacterial 12. Squire E, Favara B, Todd J (1979) Diagnosis of neonatal bacterial infection. Am Surg 69:679–686 discussion 686-677 infection: hematologic and pathologic findings in fatal and nonfatal 25. Madsen-Bouterse SA, Romero R, Tarca AL, Kusanovic JP, cases. Pediatrics 64:60–64 Espinoza J, Kim CJ, Kim JS, Edwin SS, Gomez R, Draghici S 13. Xanthou M (1970) Leucocyte blood picture in healthy full-term and (2010) The transcriptome of the fetal inflammatory response syn- premature babies during neonatal period. Arch Dis Child 45:242–249 drome. Am J Reprod Immunol 63:73–92 14. Xanthou M (1972) Leucocyte blood picture in ill newborn babies. – 26. Adkins B, Leclerc C, Marshall-Clarke S (2004) Neonatal adaptive Arch Dis Child 47:741 746 – 15. Gerdes JS (1991) Clinicopathologic approach to the diagnosis of immunity comes of age. Nat Rev Immunol 4:553 564 neonatal sepsis. Clin Perinatol 18:361–381 27. Ridge JP, Fuchs EJ, Matzinger P (1996) Neonatal tolerance 16. Josse J, Husson F (2013) Handling missing values in exploratory revisited: turning on newborn T cells with dendritic cells. Science – multivariate data analysis methods. Journal de la Société Française 271:1723 1726 de Statistique 153:79–99 28. El-Sameea ER, Metwally SS, Mashhour E, El-Bendary A, Hassan 17. Bolstad BM, Irizarry RA, Astrand M, Speed TP (2003) A compar- AM, El-Sharkawy H, El-Shennawy FA (2004) Evaluation of natu- ison of normalization methods for high density oligonucleotide ar- ral killer cells as diagnostic markers of early onset neonatal sepsis: ray data based on variance and bias. Bioinformatics 19:185–193 comparison with C-reactive protein and interleukin-8. Egypt J 18. Breitling R, Armengaud P, Amtmann A, Herzyk P (2004) Rank Immunol 11:91–102 products: a simple, yet powerful, new method to detect differential- 29. Georgeson GD, Szony BJ, Streitman K, Kovacs A, Kovacs L, ly regulated genes in replicated microarray experiments. FEBS Lett Laszlo A (2001) Natural killer cell cytotoxicity is deficient in new- 573:83–92 borns with sepsis and recurrent infections. Eur J Pediatr 160:478– 19. Leek JT, Storey JD (2007) Capturing heterogeneity in gene expres- 482 sion studies by surrogate variable analysis. PLoS Genet 3:1724– 1735